Heliyon, Volume 7, Issue 2, February 2021, e06226

Background: Hippocampus segmentation on magnetic resonance imaging is of key importance for the diagnosis, treatment decision and investigation of neuropsychiatric disorders. Automatic segmentation is an active research field, with many recent models using deep learning. Most current state-of-the art hippocampus segmentation methods train their methods on healthy or Alzheimer’s disease patients from public datasets. This raises the question whether these methods are capable of recognizing the hippocampus on a different domain, that of epilepsy patients with hippocampus resection.

New Method: In this paper we present a state-of-the-art, open source, ready-to-use, deep learning based hippocampus segmentation method. It uses an extended 2D multi-orientation approach, with automatic pre-processing and orientation alignment. The methodology was developed and validated using HarP, a public Alzheimer’s disease hippocampus segmentation dataset.

Results and Comparisons: We test this methodology alongside other recent deep learning methods, in two domains: The HarP test set and an in-house epilepsy dataset, containing hippocampus resections, named HCUnicamp. We show that our method, while trained only in HarP, surpasses others from the literature in both the HarP test set and HCUnicamp in Dice. Additionally, Results from training and testing in HCUnicamp volumes are also reported separately, alongside comparisons between training and testing in epilepsy and Alzheimer’s data and vice versa.

Conclusion: Although current state-of-the-art methods, including our own, achieve upwards of 0.9 Dice in HarP, all tested methods, including our own, produced false positives in HCUnicamp resection regions, showing that there is still room for improvement for hippocampus segmentation methods when resection is involved.

Full paper here: https://doi.org/10.1016/j.heliyon.2021.e06226

Medical Imaging with Deep Learning 2019

Hippocampus segmentation plays a key role in diagnosing various brain disorders. Nowadays, segmentation is a manual, time consuming task and considered to be the gold-standard when evaluating automated methods. For years the best performing automatic methods were multi atlas based with 80 to 85% DICE and time consuming, but machine learning methods are recently rising with promising time and accuracy performance. In thiswork, a novel method for hippocampus segmentation is presented, based on the consensus of tri-planar U-Net inspired CNNs, with some modifications based on successful CNNs of the literature, and a patch extraction technique employing data from neighbor patches. Our in-house dataset has hippocampus atrophies resulted from epilepsy surgery treatment. Our method (labeled e2dhipseg) achieves cutting edge performance of 96% DICE in ourtest data. Our method was also compared to other recent methods in the public ADNI and HARP datasets.

Full paper here: https://openreview.net/pdf?id=Sygx97DaKV

XXV Congresso Brasileiro de Engenharia Biomédica, p. 963-966, 2016

Lesions within the white matter brain are often observed on neurology and psychiatric patients, on both symptomatic and asymptomatic patients. The analysis of those lesions is a non-trivial task due to its variation on shape, location and severity. In order to help specialists to accomplish the correct diagnosis and follow-up, there are several computer-assisted tools that aim to automatically detect/segment these lesions, and to quantitatively evaluate the lesion progression over time. However, none of these tools characterize these lesions, or study its variation over time. This paper proposes a different approach to perform longitudinal study of white matter lesions that evaluate the variation of lesioned tissues over time through textural statistics, such as mean intensity and uniformity. This approach makes possible not only the quantitative longitudinal evaluation of lesions size, but also the analysis of their variation and etiology over time.

Full paper here: https://drive.google.com/drive/folders/0B543adcG1FClQ21ZaFhCUmdwMlk

Journal of Medical Imaging, 2015

Brain white matter lesions found upon magnetic resonance imaging are often observed in psychiatric or neurological patients. Individuals with these lesions present a more significant cognitive impairment when compared with individuals without them. We propose a computerized method to distinguish tissue containing white matter lesions of different etiologies (e.g., demyelinating or ischemic) using texture-based classifiers. Texture attributes were extracted from manually selected regions of interest and used to train and test supervised classifiers. Experiments were conducted to evaluate texture attribute discrimination and classifiers’ performances. The most discriminating texture attributes were obtained from the gray-level histogram and from the co-occurrence matrix. The best classifier was the support vector machine, which achieved an accuracy of 87.9% in distinguishing lesions with different etiologies and an accuracy of 99.29% in distinguishing normal white matter from white matter lesions.

Full paper here: https://doi.org/10.1117/1.JMI.2.1.014002

Pattern Recognition Letters, Volume 39, 1 April 2014, Pages 2-10

This paper presents the k-Optimum Path Forest (k-OPF) supervised classifier, which is a natural extension of the OPF classifier. k-OPF is compared to the k-Nearest Neighbors (k-NN), Support Vector Machine (SVM) and Decision Tree (DT) classifiers, and we see that k-OPF and k-NN have many similarities. This work shows that the k-OPF is equivalent to the k-NN classifier when all training samples are used as pro- totypes. Simulations comparing the accuracy results, the decision boundaries and the processing time of the classifiers are presented to experimentally validate our hypothesis. Also, we prove that OPF using the max cost function and the NN supervised classifiers have the same theoretical error bounds.

Full paper here: https://doi.org/10.1016/j.patrec.2013.08.030